Quality assurance of quantitative cardiac T1-mapping in multicenter
clinical trials
– A T1 phantom program from the hypertrophic
cardiomyopathy registry (HCMR) study
Qiang Zhang
a,⁎
,
Konrad Werys
a, Iulia A. Popescu
a, Luca Biasiolli
a, Ntobeko A.B. Ntusi
b, Milind Desai
c,
Stefan L. Zimmerman
d, Dipan J. Shah
e, Kyle Autry
e, Bette Kim
f, Han W. Kim
g, Elizabeth R. Jenista
g,
Steffen Huber
h, James A. White
i, Gerry P. McCann
j, Saidi A. Mohiddin
k, Redha Boubertakh
l,
Amedeo Chiribiri
m, David Newby
n, Sanjay Prasad
o, Aleksandra Radjenovic
p, Dana Dawson
q,
Jeanette Schulz-Menger
r, Heiko Mahrholdt
s, Iacopo Carbone
t, Ornella Rimoldi
u, Stefano Colagrande
v,
Linda Calistri
v, Michelle Michels
w, Mark B.M. Hofman
x, Lisa Anderson
y, Craig Broberg
z, Flett Andrew
aa,
Javier Sanz
ab, Chiara Bucciarelli-Ducci
ac, Kelvin Chow
ad, David Higgins
ae, David A. Broadbent
af,
Scott Semple
ag, Tarik Hafyane
ah, Joanne Wormleighton
ai, Michael Salerno
aj, Taigang He
ak, Sven Plein
al,
Raymond Y. Kwong
am, Michael Jerosch-Herold
an, Christopher M. Kramer
ao, Stefan Neubauer
a,
Vanessa M. Ferreira
a, Stefan K. Piechnik
aa
Oxford Centre for Clinical Magnetic Resonance Research, Oxford BRC NIHR, Division of Cardiovascular Medicine, Radcliffe Department of Medicine, University of Oxford, UK
b
Department of Medicine, University of Cape Town and Groote Schuur Hospital, Cape Town, South Africa
c
Cleveland Clinic, USA
d
Johns Hopkins Hospital, USA
e
Houston Methodist DeBakey Heart & Vascular Center, USA
f
Mount Sinai West Hospital; Icahn School of Medicine at Mount Sinai, USA
gDuke Cardiovascular Magnetic Resonance Center, Duke University Medical Center, USA h
Department of Radiology and Biomedical Imaging, Yale School of Medicine, New Haven, USA
i
Stephenson Cardiac Imaging Centre, Libin Cardiovascular Institute, University of Calgary, Canada
j
Department of cardiovascular sciences, University of Leicester and NIHR Leicester Biomedical Research Centre, UK
k
Inherited Cardiovascular Diseases, Barts Heart Centre, London, UK
lBarts Heart Centre, St Bartholomew's Hospital, Barts Health NHS Trust, West Smithfield, London, UK mKing's College London and Guy's and St Thomas' NHS Foundation Trust, UK
nCentre for Cardiovascular Science, University of Edinburgh, UK o
National Heart and Lung Institute, Imperial College and Royal Brompton Hospital, London, UK
p
Institute of Cardiovascular & Medical Sciences, BHF Glasgow Cardiovascular Research Centre, University of Glasgow, Glasgow, UK
q
Aberdeen Cardiovascular and Diabetes Centre, College of Life Sciences and Medicine, University of Aberdeen, UK
r
Charité, University Medicine Berlin ECRC and Helios Clinics, Berlin, Germany
s
Department of Cardiology, Robert Bosch Medical Center, Stuttgart, Germany
tDepartment of Radiological, Oncological and Pathological Sciences, Sapienza, University of Rome, Italy uUniversita' Vita Salute San Raffaele, Milan, Italy
v
Department of Experimental and Clinical Biomedical Sciences, University of Florence, Italy
w
Erasmus MC, department of cardiology, Rotterdam, the Netherlands
x
dept. Radiology and Nuclear Medicine, Amsterdam UMC location VUmc, Amsterdam, The Netherlands
y
Cardiology Clinical Academic Group, St George's University of London, UK
z
Knight Cardiovascular Institute, Oregon Health and Science University, USA
aaUniversity Hospital Southampton NHS Trust, UK abMount Sinai Hospital, New York, NY, USA ac
University of Bristol, UK
ad
Siemens Medical Solutions USA, Inc., Chicago, IL, USA
ae
Philips Electronics UK Limited, Surrey, UK
af
Biomedical Imaging Sciences Department, University of Leeds, Leeds, UK
agEdinburgh Imaging, Centre for Cardiovascular Science, University of Edinburgh, UK ahMontreal Heart Institute, Canada
aiUniversity Hospitals of Leicester NHS Trust, UK aj
University of Virginia, USA
ak
The Cardiology Clinical Academic Group (CAG), St George's University of London, St George's University Hospitals NHS Foundation Trust, UK
International Journal of Cardiology xxx (xxxx) xxx
⁎ Corresponding author.
E-mail address:qiang.zhang@cardiov.ox.ac.uk(Q. Zhang).
IJCA-29247; No of Pages 8
https://doi.org/10.1016/j.ijcard.2021.01.026
0167-5273/© 2021 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Contents lists available atScienceDirect
International Journal of Cardiology
j o u r n a l h o m e p a g e :w w w . e l s e v i e r . c o m / l o c a t e / i j c a r dPlease cite this article as: Q. Zhang, K. Werys, I.A. Popescu, et al., Quality assurance of quantitative cardiac T1-mapping in multicenter clinical trials– A T1 phantom p..., International Journal of Cardiology,https://doi.org/10.1016/j.ijcard.2021.01.026
Quality assurance Standardization Phantom study
Results: T1 temperature sensitivity followed a second-order polynomial to baseline T1 values (R > 0.996). Some phantoms showed aging effects, where T1 drifted up to 49% over 40 months. The correlation model based on ref-erence T1 and T2, developed on 1004 dedicated phantom scans, predicted ShMOLLI-T1 with high consistency (coefficient of variation 1.54%), and was robust to temperature variations and phantom aging. Using the 95% con-fidence interval of the correlation model residuals as the tolerance range, we analyzed 390 ShMOLLI T1-maps and confirmed accurate sequence deployment in 90%(70/78) of QA scans across 28 multiple centers, and categorized the rest with specific remedial actions.
Conclusions: The proposed phantom QA for T1-mapping can assure correct method implementation and protocol adherence, and is robust to temperature variation and phantom aging. This QA program circumvents the need of frequent phantom replacements, and can be readily deployed in multicenter trials.
© 2021 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
1. Introduction
CMR parametric mapping permits the quantification and spatial vi-sualization of changes in myocardial composition based on changes in T1, T2 relaxation times, and extracellular volume (ECV) [1]. Native T1-mapping has been shown to have narrow normal ranges within the same method, and sensitivity to a wide range of myocardial diseases [2,3]. Its advantages include relatively simple single breath-hold acqui-sitions, excellent reproducibility, and avoidance of gadolinium-based contrast agents.
However, T1 values depend on the protocol parameters, acquisition and reconstruction methods [4], currently requiring cumbersome es-tablishment of within-center norms [1,5]. Native T1-mapping can dif-ferentiate disease from healthy tissue in single-site studies at afixed field strength, and demonstrated reproducibility in multi-center studies under a uniform imaging set-up, but the lack of standardization limits direct comparability between centers and wider roll-out for clinical use. For clinical sites attempting to implement T1 mapping, it is often unclear how to install and validate the methods correctly before using them for clinical diagnosis, or for assessing novel therapeutics in multi-center trials. Despite SCMR consensus recommendations [1], there has been no working solution to-date for standardization– not even for a single T1-mapping technique on a single vendor platform. The use of local normal ranges and z-scores to adjust for differences have been proposed [1], but are suboptimal, as small sample size (e.g. 10–20) healthy volunteers can be prone to sampling errors [1,5]; out-liers or biases could directly affect a center's ability to detect abnormal findings to diagnose disease. While these effects decrease with larger sample sizes, re-acquiring such data frequently just to monitor stability after each method or scanner upgrade becomes prohibitively cumber-some and expensive.
Validation and quality assurance (QA) of single-method deploy-ment could be afirst step to standardize T1-mapping techniques, in-creasing the confidence of individual centers in the set-up of CMR mapping for clinical use. The large, international multicenter Hyper-trophic Cardiomyopathy Registry (HCMR) study [6] adopted a single T1-mapping method (Shortened Modified Look-Locker Inversion
Recovery, ShMOLLI [7]) to maximize comparability of the datasets to power for outcomes. This study setup provided an opportunity to collect the required phantom data to develop a QA program for standardizing T1-mapping between centers, using an original multiparametric modelling approach. The derived QA model can re-liably detect deviations from correct method implementation and protocol adherence, despite changes in phantom properties, includ-ing temperature variations [8,9] and aging effects, unlike most phan-tom solutions currently available [9,10].
2. Materials and methods 2.1. Study design
In 2013, a batch of 50 original dedicated QA phantom devices was designed for the prototype ShMOLLI T1-mapping method [7]. The manufacture details are pro-vided in Supplemental Section S1. The external appearance and arrangement of the phantom compartments are shown inFig. 1a, b, and the achieved T1 and T2 combina-tions inTable 1.
The QA program aimed to assure the detection and compensation of potential differences in T1-mapping sequence properties between sites to track T1 measure-ment stability [6]. Centers operating Siemens MR scanners and performing measure-ments free of charge (N = 28) were provided a consistent protocol developed at Oxford core lab, adapted for various Siemens software platforms. The sites were pro-vided with imaging manuals and HCMR QA protocols to perform repeated ShMOLLI-T1 acquisitions, an inversion recovery spin echo (IR-SE) acquisition and a multi-echo SE acquisition for reference T1 and T2 maps (protocol specified in Supplemental Section S2). The multi-center dataset collected from the HCMR sites served to moni-tor site-specific changes of the T1-mapping technique properties. Meanwhile phan-toms were repeatedly scanned at Oxford core lab on a MAGNETOM Avanto (1.5 T) and a MAGNETOM Trio Tim (3 T) scanners (Siemens Healthcare, Erlangen, Germany). This high-volume dataset served to investigate temperature and age de-pendencies as well as to establish the QA protocol.
The phantom scans were uploaded by individual HCMR sites to Boston core lab and then sent to Oxford core lab for data analysis between July 2014 and December 2017, when thefirst phase of the HCMR study was completed [6]. 28 HCMR sites using Siemens scanners acquired and sent 78 scans (a complete list of HCMR sites providing the phantom scans is given in Supplemental Table S1). Meanwhile, locally at Oxford core lab, we acquired 441 phantom studies between October 2013 and June 2017. Each study scanned one to three phantoms, which provided a total of 1004 phantom scans for analysis.
All reference T1 and T2 maps were calculated offline by a C++ library [11]. We analyzed the T1 and T2 using the robust mean values from the circular (15 mm
diameter) regions of interest within each of the phantom compartments (Fig. 1c), which were placed automatically by a machine learning algorithm [12]. Altogether, from each set of phantom scans, we obtained 9 sets of reference T1 (T1ref) and T2 measurements paired with 5 repeats of corresponding ShMOLLI-T1 (T1sh) for each compartment.
2.2. T1 and T2 dependencies on temperature and phantom aging
To investigate the T1 and T2 dependency on temperature, we scanned phantoms at temperatures ranging from 16 °C to 28 °C. To analyze phantom property drifts over time, four selected phantoms were scanned approximately every two weeks between Feb-ruary 2014 and June 2017, with an average of 148 studies acquired per phantom over a 40-month period. We evaluated time trends in T1ref and T2, corrected to room temperature (21 °C) using the established temperature dependencies.
2.3. T1 prediction model and T1-mapping quality assurance
We postulated that the measured cardiac ShMOLLI-T1 has a predictable dependency on T1ref and T2 in phantoms. Using the dataset acquired locally at Oxford core lab, we established a single empirical equation exploiting the inherent relationship, T1sh = Func-tion (T1ref, T2), for all T1 and T2 combinaFunc-tions. The residual offitting was calculated to es-tablish the range of agreement for the QA program. The equation was then used to predict the expected ShMOLLI-T1 from T1ref and T2, and compare it with observed ShMOLLI-T1 in the same QA scan to check the agreement. Phantom scans with ShMOLLI-T1 of all com-partments within the agreement range (the 95% confidence interval (CI) of the residual offitting; see results section) would indicate an accurate ShMOLLI sequence deployment and, therefore, QA passed. If any ShMOLLI-T1 fell outside the agreement range, the pattern of discrepancy and the scanning parameters were then further inspected, to identify po-tential sources of error. These may include artefacts, reconstruction error or protocol error, and are given a conditional pass with specific recommendations for remedial ac-tions. Failing to identify and rectify errors would suggest that the ShMOLLI sequence was deployed incorrectly, and, therefore, QA failed. The capacity of this method was vali-dated on the multi-center HCMR phantom dataset for Siemens scanners, with proof-of-principle translation demonstrated on the Philips platform.
3. Results
3.1. T1 baseline values and temperature dependency
At the time of manufacture, all 9 compartments of the 50 phantoms made as a batch had consistent ShMOLLI-T1 values with coefficient of variation within 0.85%, all measured at room temperature of ~21 °C (Table 1). In contrast to in-vivo myocardial T1 values [7], ShMOLLI-T1 values of all phantom compartments were lower at 3 T than at 1.5 T by 0.5–6.8% (all p < 0.001). Phantom T2 were typically lower at 3 T by 2.1–3.7% (all p < 0.001), with the exception of compartment #E where T2 was 1.6% higher at 3 T than at 1.5 T (p < 0.001). In compart-ments #C and #H, the T2 difference did not reach statistical significance (Table 1).
In subsequent experiments, T1ref showed clear linear dependency on temperature at both 1.5 T and 3 T (Fig. 2a). The variation of predicted T1 changes with temperature was 2.6% ± 1.5% at 1.5 T and 2.6% ± 1.2% at 3 T for all compartments, relative to the baseline T1 inTable 1. The temperature sensitivity increased with baseline T1, following closely a second-order polynomial (R2> 0.996;Fig. 2b).
3.2. Impact of phantom aging on T1
A wide range of aging effects were observed on the reference T1 measurements, with the largest seen in compartment #B in three of the four phantoms scanned repeatedly at Oxford core lab (Fig. 3). These phantoms appear to have undergone a transition, whereby the baseline T1 values had increased by ~50%. This may be due to various factors during the manufacture, particularly the amount of air within
Fig. 1. HCMR QA phantom. (a) Phantom external appearance. (b) Phantom compartment arrangement. (c) An example of T1 map in QA post-processing. Black dashed lines indicate the automatically detected ROIs. The two phantoms on either side are not part of QA; their use is recommended to assure adequate coil loading.
Table 1
Phantom chemical composition and T1/T2 relaxation times measured at room temperature within one month of manufacture. ShMOLLI T1 are average values of a batch of 50 phantoms measured at 1.5 T and 3 T. The T2 are average values offive randomly selected phantoms at 1.5 T and another five at 3 T scanned in an additional single measurement.
Phantom compartment and formulation ShMOLLI T1 [ms] (mean ± SD, N = 50) Spin-echo T2 [ms] (mean ± SD, N = 5) 1.5 T (21.3 ± 0.4 °C) 3 T (21.0 ± 0.5 °C) 1.5 T (21.0 °C) 3 T (21.0 °C) A 0.5% Agar, 0.33% Carrageenan, 0.113 mM NiCl2 2529.5 ± 14.0 2461.0 ± 7.8 275.8 ± 2.1 265.9 ± 4.8
B 0.5% Agar, 0.626 mM NiCl2 1396.2 ± 5.9 1329.4 ± 2.2 266.2 ± 4.6 259.0 ± 2.2
C Undoped 18 MΩ deionized H2O 3251.5 ± 12.5 3234.9 ± 27.8 2373.4 ± 184.3 2383.1 ± 153.9
D 1.9% Agar, 1.2 mM NiCl2 859.1 ± 2.9 804.71 ± 0.74 72.2 ± 0.7 71.4 ± 0.4
E 2% Agar, 0.77 mM NiCl2 1109.7 ± 12.5 1051.3 ± 1.9 69.6 ± 1.2 70.7 ± 0.8
F 2% Agar, 0.524 mM NiCl2 1397.8 ± 4.7 1328.0 ± 2.2 80.2 ± 1.4 78.4 ± 2.8
G 1.5% Agar, 0.1% Carrageenan, 4.5 mM NiCl2 323.1 ± 0.66 307.5 ± 0.6 72.4 ± 0.7 68.5 ± 1.1
H 3% Agar, 0.457 mM NiCl2 1428.8 ± 3.9 1368.1 ± 3.4 56.9 ± 0.4 56.5 ± 2.6
the universal container tubes and the resilience of the caps. T2 charac-teristics are provided in supplemental Fig. S4 and S5.
3.3. T1 prediction model robust to temperature variation and phantom aging Consistent with prior work, ShMOLLI-T1 generally followed a linear relationship with T1ref, with visibly larger underestimation in compart-ments with shorter T2 (Fig. 4a, red arrows). To establish a model inde-pendent of phantom age and temperature, we exploited the inherent relationships between ShMOLLI-T1 and T1ref and T2, based on the 1004 dedicated phantom datasets. We accounted for the linear depen-dency of ShMOLLI-T1 on T1ref, and an exponential dependepen-dency on T2 (coefficients calculated for 1.5 T and 3 T, respectively; full data samples in Supplemental Fig. S6). While this model significantly reduced the re-sidual differences between ShMOLLI-T1 and reference T1 methods from 7.24% to 1.64% (p < 0.001), there was a remaining small trend visible in thefit residuals. This was rectified with a third-order polynomial correction. The final multivariate correlation model predicting the ShMOLLI-T1 (cT1sh) from reference measurements were thus establi-shed; see derived equations in Suplemental S3. The model predicted c
T1sh which agreed with real T1sh with high accuracy (R2> 0.99,
Fig. 4b), despite the temperature variation and phantom aging effects described above. Analysis of the residual errors of the ShMOLLI predic-tion model allowed the establishment of the 95% confidence interval CI = ±3.12%, and the 99.7% confidence interval CI = ±5.32%, robust to temperature and aging confounders. The 95% CI was used as the tol-erance range for the QA (full data samples in Supplemental Fig. S7).
3.4. Clinical application: T1-mapping quality assurance
We received 94 phantom scans: 78 from 28 Siemens sites, 15 from 6 Philips sites, and 1 from a General Electric (GE) site (complete list of sites in Supplemental Table S1). The vendor and scan distribution reflected the resources available and local feasibility during this study, which precluded fair head-to-head inter-vendor comparisons. We were able to perform QA and present thefindings in 78 scans (390 ShMOLLI-T1 maps) from the 28 Siemens sites (Table 2andFig. 5). We exemplified the need for further work for inter-vendor application of the QA model (Fig. 5f), subject to sufficient datasets acquired on those MR systems.
3.5. QA passed
QA was passed if the dataset from a session contained at least one accurate ShMOLLI-T1 acquisition for all 9 compartments. 34 scans from 15 sites gave ShMOLLI-T1 values within the prescribed 95% CI range of the expected ShMOLLI-T1, confirming accurate ShMOLLI de-ployment (Fig. 5a). Three scans from 3 sites showed departure of ShMOLLI-T1 values in individual phantom compartments, with the rest being within the agreement range. We identified the sources of these outliers as imaging artefacts in the reference T1-maps (Fig. 5b). These 3 scans were considered to have passed the QA, as the source of discrepancies was clearly outside the cardiac T1-mapping technology, and the ShMOLLI-T1 values of the rest of the compartments were within the agreement range.
Fig. 2. Temperature sensitivity of reference T1 in the Oxford core lab dataset at 1.5 T and 3 T. (a) T1 temperature dependency. Temperature sensitivity coefficients (ΔT1/Δt, ms/°C) are provided to the right of the graph, prefixed with the compartment ID. Regression lines are omitted for clarity. (b) T1 temperature sensitivity coefficients (Y-axis) follow a second-order polynomial to baseline T1 values (X-axis).
3.6. Common error 1: Lower T1s due to inadequate waiting time
A further 26 scans from 16 sites provided at least one accurate ShMOLLI acquisition but with a characteristic pattern of greater under-estimation of ShMOLLI-T1 values in longer T1 compartments in other acquisitions of the scan (Fig. 5c), suggesting incomplete recovery of magnetization. This was likely caused by inadequate waiting time after the previous image acquisition, frequency adjustment, or shim-ming adjustment. We reproduced this T1 underestimation at Oxford core lab with dedicated experiments, and studied the impact of incom-plete magnetization recovery on the measured T1. The experimental details are provided in Supplemental Section S4. This error may be rec-tified by reminding the operators to wait at least 10 s in between T1-map acquisitions.
3.7. Common error 2: failed in-line reconstruction of T1-mapping We detected incorrect in-line ShMOLLI-T1 reconstruction in 7 addi-tional scans from 5 sites (Fig. 5d, gray circles). We were able to recon-struct and salvage T1-maps offline using raw T1-weighted images, and restored the accurate T1 values (Fig. 5d, blue points). This indicated cor-rect T1-weighted image acquisition but inaccurate inline reconstruction on the MR system. Therefore, these sites passed QA conditionally on offline reconstruction, but required redeployment of the T1 sequence. 3.8. QA failed
Six scans did not provide T1 measurements falling wholly within the prescribed tolerance limit, and therefore did not pass QA (Fig. 5e). A further 2 scans from 2 sites contained no reference T1 or T2 acquisi-tions, and thus the QA was not performed. Similarly, incompatible
measurement protocols prevented a convincing inter-vendor valida-tion, with one example shown as a proof of principle (Fig. 5f).
4. Discussion
In this work, we have established a novel approach for quality assur-ance (QA) that is independent of the actual physical properties of the phantoms, bypassing the exhibited phantom sensitivities to tempera-ture variations and aging. We have demonstrated how to use this QA to verify correct implementation of T1-mapping methods to within a prescribed tolerance range across multiple scanners and magnetic field settings; signature patterns of departure also identified common errors for actionable remediations.
4.1. Inter-method and inter-vendor applications
We have illustrated how to achieve this phantom QA solution using a single T1-mapping method, but the solution can be deployed to other mapping techniques. The HCMR Consortium had chosen a single T1-mapping method (ShMOLLI [7]) to maximize comparability of the datasets to power for outcomes. The ShMOLLI T1-mapping sequence possesses three characteristics which made this work possible within 5 years: (1) method stability and full accountability of its quantitative characteristics, such that it was feasible to compare it on a wide range of platforms within a single vendor setting; (2) heart-rate indepen-dence, alleviating the need to vary heart rate as part of the QA process; (3) ShMOLLI is a single universal technique suitable for measuring a wide range of T1 values (whether short or long), thus obviating the need for developing separate QA models, each for a specific MOLLI sub-variant for short or long T1s.
Fig. 3. The varied appearance of age-related drifts in reference T1 (T1ref) in individual phantom compartments observed over a period of 40 months. All T1ref values are corrected to room temperature. Colors represent the four phantoms investigated.
Fig. 4. ShMOLLI T1 (T1sh) against reference T1 (T1ref) in the Oxford core lab dataset before T2 correction (a) and after T2 correction (b) at 1.5 T and 3 T. (a) The visible deviations (red arrows) were driven predominantly by T2 effects. (b) The multivariate correlation model including T2 effects predicts T1sh with high accuracy (R2
> 0.99). (For interpretation of the references to colour in thisfigure legend, the reader is referred to the web version of this article.)
Table 2
QA results of 78 Siemens scans from 28 sites with proposed outcomes and actions.
QA Results Description No. scans
(sites)
Action recommended
1. Passed All T1 maps in the scan provided ShMOLLI-T1 (T1sh) values within the agreement range with expected cT1sh (Fig. 5a).
34 scans (15 sites)
QA passed.
No further action required Disagreement between T1sh and cT1sh in one or more individual compartments; the rest were in
the agreement range. The source of error can be identified as image artefacts not linked to cardiac T1 maps (Fig. 5b).
3 scans (3 sites)
QA passed.
Consider technical investigation. 2. Warnings Underestimated T1sh in individual acquisitions. At least one acquisition is within the agreement
range. Incomplete recovery of longitudinal magnetization in individual acquisitions (Fig. 5c).
26 scans (16 sites)
QA passed, with warning of possible protocol adherence problems 3. Conditional T1sh values outside the agreement range; Source of disagreement caused by T1 mapfitting
without ShMOLLI conditionalfitting reconstruction, but accurate T1sh values were successfully restored offline (Fig. 5d).
7 scans (5 sites)
QA conditional on offline reconstruction. Require re-deployment of T1 sequence 4. Failed T1sh values outside the agreement range; unable to identify source of error. Unable to restore
accurate T1sh values offline (Fig. 5e).
6 scans (3 sites)
QA not passed. Technical investigation required
QA could not be performed due to missing reference T1 or T2 sequences in the scan. 2 scans (2 sites)
Incomplete scan. Check protocols and repeat QA
platform. Although the general approach can be applied across vendors, the investigators' experience indicates that translation to other vendors and methods will require significant further technical investment, to address the large intervendor differences between documented sim-ilar methods [10]. The authors are open to collaborative requests to employ the proposed QA to improve the T1-mapping consistency in multi-vendor setting. In future, the described QA should be either im-plemented directly on the scanner or be provided as a prompt external service, so that end-users can assure method stability with the same fre-quency as the usual scanner QA. This QA is designed to test correct tech-nical deployment of methods; operator compliance with the acquisition protocols and standardization of image processing [13] are required to minimize introduction of errors.
5. Conclusions
We presented the development of a practical MR phantom QA pro-gram for accurate and comparable T1 measurements for use in a large multi-center setting. The QA model is robust to phantom aging and am-bient temperature variations, circumventing the need for manual tem-perature corrections and frequent phantom replacements. This provides an immediate and economical solution to verify correct T1-mapping method implementation, and identify common errors for re-mediation. The proposed QA program paves the way to widespread adoption of T1-mapping into routine clinical practice.
Funding
This work is funded by British Heart Foundation (BHF) project grant PG/15/71/31731 (QZ, KW, SKP) and National Institute for Health Re-search (NIHR) Oxford Biomedical ReRe-search Centre at The Oxford Uni-versity Hospitals (IAP, SKP, VMF and SN).
Declaration of Competing Interest
SKP has patent authorship rights for U.S. patent US20120078084A1. Systems and methods for shortened Look Locker inversion recovery (Sh-MOLLI) cardiac gated mapping of T1. Granted March 15, 2016. IP is managed by Oxford University Innovations; the license exclusively transferred to Siemens Healthcare.
QZ, SKP, KW, IAP, VMF have authorship rights for pending patent PCT/GB2020/051189. A method for identity validation and quality as-surance of quantitative magnetic resonance imaging protocols. Filed
Supplementary data to this article can be found online athttps://doi. org/10.1016/j.ijcard.2021.01.026.
References
[1] D.R. Messroghli, J.C. Moon, V.M. Ferreira, L. Grosse-Wortmann, T. He, P. Kellman, et al., Clinical recommendations for cardiovascular magnetic resonance mapping of T1, T2, T2* and extracellular volume: a consensus statement by the Society for Cardiovascular Magnetic Resonance (SCMR) endorsed by the European Association for Cardiovascular Imaging (EACVI), J. Cardiovasc. Magn. Reson. 19 (1) (2017) 75.
[2] E.B. Schelbert, D.R. Messroghli, State of the art: clinical applications of cardiac T1 mapping, Radiology 278 (3) (2016) 658–676.
[3] J.M. Liu, A. Liu, J. Leal, F. McMillan, J. Francis, A. Greiser, et al., Measurement of myo-cardial native T1 in cardiovascular diseases and norm in 1291 subjects, J. Cardiovasc. Magn. Reson. 19 (2017) 74.
[4] I.A. Popescu, K. Werys, Q. Zhang, H. Puchta, E. Hann, E. Lukaschuk, et al., Standardi-zation of T1-mapping in cardiovascular magnetic resonance using clustered struc-turing for benchmarking normal ranges, Int. J. Cardiol. (2020).
[5] J.C. Moon, D.R. Messroghli, P. Kellman, S.K. Piechnik, M.D. Robson, M. Ugander, et al., Myocardial T1 mapping and extracellular volume quantification: a Society for Car-diovascular Magnetic Resonance (SCMR) and CMR Working Group of the European Society of Cardiology consensus statement, J. Cardiovasc. Magn. Reson. 15 (1) (2013) 92.
[6] C.M. Kramer, E. Appelbaum, M.Y. Desai, P. Desvigne-Nickens, J.P. DiMarco, M.G. Friedrich, et al., Hypertrophic cardiomyopathy registry: the rationale and design of an international, observational study of hypertrophic cardiomyopathy, Am. Heart J. 170 (2) (2015) 223–230.
[7] S.K. Piechnik, V.M. Ferreira, E. Dall’Armellina, L.E. Cochlin, A. Greiser, S. Neubauer, et al., Shortened Modified Look-Locker Inversion recovery (ShMOLLI) for clinical myocardial T1-mapping at 1.5 and 3 T within a 9 heartbeat breathhold, J. Cardiovasc. Magn. Reson. 12 (2010) 69.
[8] W.D. Zech, N. Schwendener, A. Persson, M.J. Warntjes, C. Jackowski, Temperature dependence of postmortem MR quantification for soft tissue discrimination, Eur. Radiol. 25 (8) (2015) 2381–2389.
[9] G. Captur, P. Gatehouse, K.E. Keenan, F.G. Heslinga, R. Bruehl, M. Prothmann, et al., A medical device-grade T1 and ECV phantom for global T1 mapping quality assurance-the T-1 mapping and ECV standardization in cardiovascular magnetic resonance (T1MES) program, J. Cardiovasc. Magn. Reson. 18 (2016) 58.
[10] G. Captur, A. Bhandari, R. Brühl, B. Ittermann, K.E. Keenan, Y. Yang, et al., T(1) map-ping performance and measurement repeatability: results from the multi-national T (1) mapping standardization phantom program (T1MES), J. Cardiovasc. Magn. Reson.: Off. J. Soc. Cardiovasc. Magn. Reson. 22 (1) (2020) 31.
[11] K. Werys, I. Dragonu, Q. Zhang, I. Popescu, E. Hann, H. Puchta, et al., Total mapping toolbox (TOMATO): an open source library for cardiac magnetic resonance para-metric mapping, SoftwareX 11 (2020) 100369.
[12] L. Biasiolli, J.A. Noble, M.D. Robson, Multicontrast MRI registration of carotid arteries in atherosclerotic and normal subjects, Medical Imaging 2010: Image Processing, In-ternational Society for Optics and Photonics, 2010.
[13]V. Carapella, H. Puchta, E. Lukaschuk, C. Marini, K. Werys, S. Neubauer, et al., Stan-dardized image post-processing of cardiovascular magnetic resonance T1-mapping reduces variability and improves accuracy and consistency in myocar-dial tissue characterization, 298, 2020 128–134.
Fig. 5. Phantom QA of T1-mapping for 6 example CMR centers. In each panel, the top graph shows observed ShMOLLI T1 (T1sh) and the bottom graph the residuals (y-axis) displayed against the expected cT1sh (x-axis). (a) QA passed as all within 95% CI (green range). (b) QA passed with artefacts in individual compartment(s). (c) QA passed with warning. Underestimated T1sh in individual acquisitions. (d) QA conditional on T1sh offline reconstruction. Inline reconstruction failed (gray circles). (e-f) QA failed due to presence of patterns and variability of the observed residuals reaching outside the tolerance range. (For interpretation of the references to colour in thisfigure legend, the reader is referred to the web version of this article.)
